Chart Creator Agent

The Chart Creator Agent scans markdown documents for tables, builds a BAR or LINE chart from the data in each one worth charting, and returns the documents with each rendered chart placed directly after its source table. Use it when a document-producing step should turn tabular data into a visual instead of leaving the reader to parse rows and columns.

The Chart Creator Agent belongs to the Generation group in the agent library. It is a preset over the Custom Agent type: a model-backed agent that ships with its Instructions already written and the code interpreter Tool switched on. It is not the externally dispatched Custom Agent (External) type, and there is no custom agent ID to pin.

When to use it

Use the Chart Creator Agent when a workflow produces markdown containing tabular data and you want that data visualised inline:

• Turn a table of monthly figures into a bar or line chart directly beneath it.
• Add visual summaries to a generated report without hand-authoring chart markup.
• Run downstream of a Document Writer Agent or Data Extractor Agent that produced the source tables.

The agent draws BAR and LINE charts only. Pie, scatter, stacked, area, and heatmap charts are explicitly excluded. It also leaves a table uncharted when neither shape clarifies the data.

How it works

The agent's Instructions walk the model through a three-step pipeline for every table it finds:

• Typecast the columns. Infer whether each column is datetime, number, string, or categorical, treating values such as -, N/A, null, and blank as missing.
• Decide what to draw. A BAR chart compares a numeric measure across categorical values; a LINE chart shows a measure changing over a datetime or ordered numeric axis.
• Render with the code interpreter. The agent writes a Python script using pandas, matplotlib, and seaborn, and saves each chart as a PNG artifact.

Because rendering runs through the code interpreter Tool, that Tool is switched on by the preset and is not meant to be turned off.

Inputs

{
  "texts_with_markdown": [
    "# Q1 Results\n\n| Month | Revenue |\n|---|---|\n| Jan | 120 |\n| Feb | 150 |\n| Mar | 175 |\n"
  ],
  "runtime_instructions": "Chart revenue only. Sort months chronologically."
}

Controlling the charts

The styling defaults live as editable prose inside the agent's Instructions, not as separate settings fields. Open the agent and edit its Instructions to change any of them:

The Instructions carry a dedicated plotting-instructions section for saying which tables and columns to chart, and in what order and emphasis. For guidance that changes per run, use runtime_instructions instead of editing the Instructions.

Output

The agent returns the original markdown as text, with each chart's image link inserted immediately after the table it was built from. Uncharted tables and all surrounding text are returned unchanged.

# Q1 Results

| Month | Revenue |
|---|---|
| Jan | 120 |
| Feb | 150 |
| Mar | 175 |

![q1_revenue](https://.../q1_revenue.png)

A downstream agent reads the result via {{<node_id>.output}}. Because the charts are returned as artifact links rather than embedded data, whatever renders the markdown resolves the images.

What's next

• Custom Agent — the agent type this preset is built on, and how its Instructions and Tools work.
• Document Writer Agent — a common upstream source of the tables this agent charts.
• Agent Reference overview — the agent-type model, catalog, and shared execution engine.